• Semichem.com
  • Semichem.com
  • About Us
  • Contact
  • Support
• AMPAC™
  • AMPAC™
  • Features
  • Platforms
  • Pricing
  • Support
• CODESSA™
  • CODESSA™
  • Features
  • Platforms
  • References
  • Pricing
  • Support
• AGUI™
  • AGUI™
  • Features
  • Platforms
  • Pricing
  • Support
• Download
  • Download
  • Documentation
  • Login
  • Register

Semichem, Inc. — Providers of Solutions for Computational Chemistry

Comparison¹
AMPAC 8 and MOPAC 2002

1. For single point energy calculations and optimizations, one AMPAC SCF cycle was about twice as fast as a MOPAC SCF cycle.


2. AMPAC 1SCF calculations frequently took 20-30% more SCF cycles for convergence than MOPAC. However, since AMPAC's SCF cycles are faster, the total calculation typically took only 2/3 the time for MOPAC.


3. For optimizations² run on polygycine chains (4-24 units long) in an ?-helix configuration, AMPAC succeeded in all cases while MOPAC failed to fully converge in 50% for the cases³. For cases where MOPAC succeeded, it was 5 to 8 times slower than the corresponding AMPAC calculation! MOPAC calculations took 2 to 5 times as many geometry steps as AMPAC as well as being 1/3 slower for each SCF.


4. Two additional large biological molecules (130 and 173 atoms respectively) were optimized by AMPAC (using TRUSTE) and MOPAC (using EF) with GNORM=1.0 for both. AMPAC took 51 and 3 geometry steps respectively, while MOPAC required 580 and 460 steps starting from the same geometry. For these two cases, AMPAC was 12 times (73 versus 927 sec) faster and 123 times (16 versus 1960 sec) faster than MOPAC.


5. For frequency calculations (FORCE), AMPAC was typically 1.5 to 2.5 times faster than the corresponding MOPAC calculation⁴. For special and uncommon high symmetry cases, MOPAC takes advantage of symmetry to speed up the calculation⁵.


6. For stationary point characterization, AMPAC provides LFORCE for which there is no parallel in MOPAC. LFORCE just computes the first few frequencies (lowest), which are sufficient for characterizing stationary points. For these cases, AMPAC with LFORCE was 10-20 times faster than MOPAC with FORCE for molecules with around 200 atoms. Since this speed up is proportional to the size of the system, this speed advantage will be more pronounced for larger molecules.


7. Configuration interaction (C.I.) is used to compute the properties of exceited states and open shell systems. C.I. AMPAC is profoundly faster than C.I. in MOPAC for identical systems and offers far greater functionality. For energy points on smaller systems (20-30 atoms with C.I.=8) or larger systems (57 atoms with C.I.=4), AMPAC was easily 100-500 times faster than the corresponding MOPAC calculation. When optimizing these same structures⁶, AMPAC was 1,000 to 3,000 times faster. Again, this speed advantage will be more pronounced for larger molecules.


8. For transition state(TS) location, AMPAC has TRUSTG while MOPAC uses the TS algorithm. To test the performance of these two algorithms, 13 different and varied reactions were studied. In each case, the optimization was started from a reasonable guess near the correct TS. Both TRUSTG and TS were comparable in reliability for finding the TS. However, TRUSTG was faster than MOPAC’s TS in 10 out of 13 cases, in which case it ranged from 1.2 to 6.7 times faster. In the two cases that TS was faster, it was faster by no more than a factor of 2.


9. In addition to TRUSTG for locating transition states, AMPAC also has LTRD. LTRD uses a full second-derivative matrix (Hessian) to define its search direction, and can handle cases that will fail with other approaches. MOPAC has no comparable capability.


10. Using the same set of 13 reactions but starting from the reactants and products, CHN in AMPAC and SADDLE in MOPAC were compared in the ability to locate the accompanying transition state. CHN succeeded in finding the transition state in all 13 cases. MOPAC’s SADDLE algorithm was successful for only 3 cases. In 9 of the cases, SADDLE stopped prior to satisfying the gradient norm criteria (GNORM=1.0) and in another case the job never completed. For the 12 jobs that did finish, CHN was 4 to 100 times faster than SADDLE.

¹All calculations were performed on the same dual processor Xeon system running Windows XP Pro.
²Default optimizers used: AMPAC TRUSTE, MOPAC EF.
³Calculations were run using the default optimizers with GNORM=1.0.
⁴Symmetry not considered.
⁵For example, a factor of 7-8 speed up for Td symmetry.
⁶Using default optimizers and GNORM=1.0